A satellite-observed climatology of global temporal autocorrelations can be related to aerosol lifetimes

Schutgens, Nick; Andrews, Elisabeth J.; Arola, Antti; Bhatti, Yusuf; Fu, Guanliang; Hasekamp, Otto; Kolmonen, Pekka; Lipponen, Antti; Mielonen, Tero; Reygare, Leighton; Sayer, Andrew M.

doi:10.5194/egusphere-2026-2656

Preprints

https://doi.org/10.5194/egusphere-2026-2656

Preprints

29 May 2026

| 29 May 2026

A satellite-observed climatology of global temporal autocorrelations can be related to aerosol lifetimes

Nick Schutgens, Elisabeth J. Andrews, Antti Arola, Yusuf Bhatti, Guanliang Fu, Otto Hasekamp, Pekka Kolmonen, Antti Lipponen, Tero Mielonen, Leighton Reygare, and Andrew M. Sayer

Abstract. Temporal autocorrelations of aerosol are often reported, but poorly understood. We use simple box models, Perturbed Parameters Ensembles of global aerosol models, AEROCOM (AEROsol Comparison of Observations and Models) simulations as well as satellite and AERONET (AErosol RObotic NETwork) observations to study temporal autocorrelations in aerosol optical depth (AOD). In particular, we present the first global climatology of observed temporal autocorrelations.

We develop a conceptual model for autocorrelations and relate them to important timescales, in particular lifetimes. We identify aerosol processes that affect autocorrelations and find autocorrelations provide information independent from yearly AOD, in particular on deposition processes. It is possible to estimate temporal autocorrelations in AOD from satellite observations by sensors like MODIS (MODerate resolution Imaging Spectroradiometer) or POLDER (POLarization and Directionality of the Earth’s Reflectances).

In our unique global climatology of observed temporal autocorrelations, regional variation is significant. Over remote oceans, the autocorrelation after 6 days tends to be low (∼ 0.2 or lower) but it is quite high in tropical outflow regions (∼ 0.5). Over land, it varies considerably, from 0.2 to 0.7. These spatial variations are much larger than observed year-to-year variation.

AEROCOM models often significantly overestimate autocorrelations. This suggests that loss processes are underestimated and/or contributions from seasonal sources are overestimated, which should have a marked impact on aerosol forcing estimates. Autocorrelations offer a new way to understand aerosol processes and evaluate models. Autocorrelations can be derived from existing observational datasets, for example surface black carbon mass concentrations or cloud condensation nuclei.

Received: 08 May 2026 – Discussion started: 29 May 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 2860 KB)

Supplement (1654 KB)

Download & links

Nick Schutgens, Elisabeth J. Andrews, Antti Arola, Yusuf Bhatti, Guanliang Fu, Otto Hasekamp, Pekka Kolmonen, Antti Lipponen, Tero Mielonen, Leighton Reygare, and Andrew M. Sayer

Status: final response (author comments only)

RC1:
'Comment on egusphere-2026-2656', Anonymous Referee #1, 29 Jun 2026
The study by Schutgens et al. is based on the temporal autocorrelation of AOD (a proxy for aerosol lifetime under certain conditions) and the key drivers of these autocorrelations. The authors provide a theoretical explanation of the relationship between the autocorrelation timescale and lifetime. By comparing temporal AOD autocorrelations from several satellite products with AERONET observations, they demonstrate that satellite snapshot data can be a useful tool for model evaluation. They then assess the performance of AEROCOM model simulations in reproducing AOD autocorrelations using satellite-based estimates. I find the methodology to be sound, and the uncertainties and limitations are discussed and quantified adequately. I have only a few minor comments aimed at improving the clarity of the manuscript. I recommend acceptance after the following comments have been addressed.
It is not mentioned how the fractional variance (shown in Fig. 4) is calculated for AOD and, more importantly, for the temporal autocorrelation at a lag of 6 days (ρ₆). It would also be useful to show how much variance (in absolute terms) is explained collectively by all of the contributing components for AOD, perhaps in a supplementary figure.

Fig 4a: The legend for EMI_DMS is not accurate.

Lines 252-253: It is surprising to see sea salt and DMS contributing substantially to the AOD variance over a predominantly continental region such as Australia. Could the authors comment on the physical reasons for this pattern?

Lines 247-249: The statement, "The first thing to notice is that the same processes are important for AOD and ρ₆ variance, although their relative contributions can be quite different" appears to contradict the discussion that follows, where the dominant contributors to AOD variance and ρ₆ variance differ in most regions. This statement seems more appropriate for the UKESM ensemble results shown in Figure 6.

Figure 8: It is not clear how the different lag values given in the caption should be interpreted, since this figure compares AOD rather than autocorrelation. In addition, the mean values reported in the legend of each panel appear to be significantly lower than suggested by the scatter points. Please verify these values. The same issue with mean values appears to be present in Figures 9 and 11.

Figure 8: Please state explicitly in the figure caption that the panels show comparisons for different satellite products. It would also be helpful to mention that the satellite products are based on 1° × 1° averages located close to or overlapping the AERONET sites. Also, please provide a list of the AERONET stations used in these comparisons in the supplementary.

The description of the AERONET sampling used to compute the autocorrelations shown in Figure 9 is missing. Were only the collocated AERONET observations corresponding to the satellite overpass times used (i.e., approximately one observation per day)? If so, the AERONET-derived autocorrelation coefficients may depend on the satellite overpass time and could therefore differ among the various panels of Figure 9. Please clarify.

Lines 349-354: Is the information discussed here presented elsewhere, for example in a supplementary figure or table? If so, please provide the appropriate reference.

Section 8: A useful addition would be a direct comparison of AOD between the models and the satellite products. Since the manuscript already discusses which processes dominate the variance in modelled AOD (Figures 4 and 6), such a comparison would help explain the differences in the estimated autocorrelations between the models and observations.

Lines 440-445: Given the discussion in Section 2.6, could the authors comment more explicitly on how the inferred autocorrelation timescales relate to aerosol lifetimes? Establishing this connection would help strengthen the physical interpretation of the results.
Citation: https://doi.org/10.5194/egusphere-2026-2656-RC1
- AC1: 'Reply on RC1', Nick Schutgens, 17 Jul 2026
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-2656/egusphere-2026-2656-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2026-2656-AC1
RC2:
'Comment on egusphere-2026-2656', Anonymous Referee #2, 25 Sep 2026
This study by Schutgens et al presents a comprehensive satellite climatology of temporal autocorrelations of AOD in different regions, and evaluates AEROCOM models against these observations. It also demonstrates that autocorrelation and AOD variance within large regional domains frequently responds differently in PPEs, showing that autocorrelation as a metric provides unique insight into aerosol processes. The study is interesting and provides a novel method for evaluating modelled aerosol processes against commonly-used and widely-available data products. For that reason, the study should be published. However, there are some things I want to discuss:
1) The effect of large spatial aggregation on measures of both autocorrelation and variance: In section 3, the authors argue that the average outflow time from continents exceeds established average aerosol lifetimes, and that loss from the domain would therefore be dominated by deposition. However, sources within the regions are not evenly distributed, and industrial cities on the leeward sides of continents may contribute considerably to regional AOD, and regional AOD variance, but less to autocorrelation, due to outflow. Autocorrelations in a eulerian framework would primarily capture the atmospheric lifetimes of aerosol emitted on the windward side of a domain, which may not be representative of the entire region. Similarly, to the extent that aerosol processes within different parts of the domain are independent of one another, aggregating over increasingly large domains would tend to decrease AOD variance, just due to the aggregation, while I would expect autocorrelation to increasingly depend on the persistence of only the most extreme emissions events. Their demonstration that the aerosol's decorrelation time reflects its lifetime only when the source itself has low autocorrelation is useful. I would recommend trimming much of lines 195-205 and then following this with a brief discussion of what such relationships mean when 'source' is actually a composite of many different sources, with potentially different amplitudes, persistences and distances from the domain edge.
2) Related to the comment above, “lifetimes” may not be the most conclusive or most interesting aspect of this work: the title “...can be related to aerosol lifetimes” sounds hedged, and indeed I think it is. However, the authors show that autocorrelation as a diagnostic provides interesting information about the temporal characteristics of aerosol loadings, even in the absence of a clear deposition dependency, which is also what the authors focus on in their conclusion. I suggest the authors elaborate on whether different combinations of low/high variance and low/high autocorrelation correspond to distinct temporal patterns (e.g., short-lived episodes, slow undulations); a 2×2 table could make this comparison easily. In section 4 it would be interesting to note if a particular parameter pushes a region towards one “type” of temporal pattern. Between this and sections 7 and 8, the authors establish that autocorrelation is a useful and novel diagnostic providing a more complete picture of the temporal characteristics of aerosol than variance alone; I suggest that the authors consider this their central conclusion, and alter the introduction, abstract, and title accordingly.

Minor comments:
throughout the work, change “lifetimes” to “local lifetimes" or “domain lifetimes”

line 150 — mention whether the L2 data are screened for quality — if not, might this bias the sample? if so, could the screening introduce sampling biases?

lines 157 — worth mentioning that in some areas data availability and data quality varies by season

line 155 — may be helpful to compare this colocation window to those used by validation studies (e.g. Levy et al., 2013 https://doi.org/10.5194/amt-6-2989-2013. Munchak et al., 2013 https://doi.org/10.5194/amt-6-1747-2013, Virtanen et al., 2018 https://doi.org/10.5194/amt-11-925-2018)

sections 2.1, 2.2, and 2.6: somewhere the manuscript should acknowledge that AOD also depends on environmental factors like humidity, as well as particle aging, and spatially or temporally varying retrieval biases, which can affect autocorrelations calculations even in the absence of mass loss from the domain.

lines 237-238 — Following my comment above, I'm not convinced the manuscript has demonstrated that outflow's contribution is minor

lines 346-348 — they could correlate if retrievals are affected by cloud-adjacency, which would produce 3D artifacts that would depend on the SZA; worth mentioning in section 2.5, alongside discussions of retrieval quality or selection biases arising from quality screening (Smith et al., 2025 https://doi.org/10.5194/acp-25-14333-2025)

lines 367-371 — I suggest the authors elaborate on what a high or low autocorrelation implies about the shape of the underlying time series, and the same for the models' overestimated autocorrelations (line 381)
Citation: https://doi.org/10.5194/egusphere-2026-2656-RC2

Nick Schutgens, Elisabeth J. Andrews, Antti Arola, Yusuf Bhatti, Guanliang Fu, Otto Hasekamp, Pekka Kolmonen, Antti Lipponen, Tero Mielonen, Leighton Reygare, and Andrew M. Sayer

Supplement

https://doi.org/10.5194/egusphere-2026-2656-supplement

Nick Schutgens, Elisabeth J. Andrews, Antti Arola, Yusuf Bhatti, Guanliang Fu, Otto Hasekamp, Pekka Kolmonen, Antti Lipponen, Tero Mielonen, Leighton Reygare, and Andrew M. Sayer

Viewed

Total article views: 324 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
193	92	39	324	58	30	29

HTML: 193
PDF: 92
XML: 39
Total: 324
Supplement: 58
BibTeX: 30
EndNote: 29

Views and downloads (calculated since 29 May 2026)

Month	HTML	PDF	XML	Total
Jun 2026	58	20	3	81
Jul 2026	49	31	3	83
Aug 2026	30	12	15	57
Sep 2026	56	29	18	103

Cumulative views and downloads (calculated since 29 May 2026)

Month	HTML	PDF	XML	Total
Jun 2026	58	20	3	81
Jul 2026	49	31	3	83
Aug 2026	30	12	15	57
Sep 2026	56	29	18	103

Viewed (geographical distribution)

Total article views: 289 (including HTML, PDF, and XML) Thereof 289 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 30 Sep 2026

Short summary

Aerosol are fine particles in the air that affect public health and the climate. We develop a new method to study aerosol through patterns (called autocorrelations) in their time-series. We show that autocorrelations are linked to aerosol deposition processes and contain information on the important but uncertain aerosol lifetime. Autocorrelations can be observed from satellites and we provide the first ever global climatology of aerosol autocorrelations.


Total:	0
HTML:	0
PDF:	0
XML:	0